Hi Yohei,

I am not sure what exactly do you mean by "recommended/correct way"
but in case of TUM we normally define new types on per package base,
e.g. for our "pcl_cloud_alogs" package we simply have definitions in
"pcl_cloud_algos_point_types.h and pcl_cloud_algos_point_types.hpp"
files and just use them as every other point type defined in pcl.

D.

On Tue, Jan 25, 2011 at 4:23 AM, Yohei Kakiuchi
<youhei@jsk.t.u-tokyo.ac.jp> wrote:
> I confirmed that a user-defined point type described in
> http://www.ros.org/wiki/pcl_ros worked well for our purpose.
>
> I have a question for using a user-defined point type.
> PointXYZRGB defined in point_types.hpp is using PCL_ADD_POINT4D,
> EIGEN_ALIGNE16.
> I tested simple user-defined class not using EIGEN_ALIGN16 and
> a user-defined class added 'category' to PointXYZRGB.
> Both classes worked well.
> Which is the recommended / correct way for using a user-defined class?
>
> * test code
>
> ---
> namespace my_ns
> {
> #if 0
>  struct PointXYZRGBClass
>  {
>   float x;
>   float y;
>   float z;
>   float rgb;
>   uint32_t category;
>  };
> #else
>  struct PointXYZRGBClass
>  {
>    PCL_ADD_POINT4D;  // This adds the members x,y,z which can also be
> accessed using the point (which is float[4])
>    float rgb;
>    uint32_t category;
>    EIGEN_MAKE_ALIGNED_OPERATOR_NEW
>  } EIGEN_ALIGN16;
> #endif
> } // namespace my_ns
>
> POINT_CLOUD_REGISTER_POINT_STRUCT
> (my_ns::PointXYZRGBClass,
>  (float, x, x)
>  (float, y, y)
>  (float, z, z)
>  (float, rgb, rgb)
>  (uint32_t, category, category)
>  );
>
> 2011/1/24 Kei Okada <k-okada@jsk.t.u-tokyo.ac.jp>:
>>> Frequently, these problems are better solved by republishing some or
>>> all of the original points as a single PointCloud2 with a
>>> categorization field. The data can be sorted on that field for
>>> efficient access, sometimes that occurs naturally as a result of your
>>> algorithm, anyway.
>>
>> I see, that's seems good solution.
>>
>> Just to be clear, your suggestion is to make PointCloud2 data as followings.
>>    sensor_msgs::PointCloud2 cloud2_;
>>    cloud2_.fields.resize( 5 );
>>    cloud2_.fields[0].name = "x";
>>    cloud2_.fields[1].name = "y";
>>    cloud2_.fields[2].name = "z";
>>    cloud2_.fields[3].name = "rgb";
>>    cloud2_.fields[3].name = "label" or "category";
>> if this is correct, is there a way to convert from/to pcl::Point*** ??
>> Usually, we will write program using pcll:PointCloud<pcl::PointXYZRGB>...
>>
>>
>> On Sun, Jan 23, 2011 at 2:43 AM, Jack O'Quin <jack.oquin@gmail.com> wrote:
>>> On Fri, Jan 21, 2011 at 7:32 AM, cmuell2s
>>> <christian.mueller@smail.inf.fh-bonn-rhein-sieg.de> wrote:
>>>> Hello all,
>>>>
>>>> I stuck at a small problem. I would like to publish a vector/array of
>>>> pcl::PointCloud<pcl::PointXYZ>.
>>>> I couldn't find a sensor_msg or std_msg which would do it, did I
>>>> overlooked sth?
>>>
>>> This comes up frequently. In many cases it seems the obvious solution.
>>>
>>> Often, problems become apparent on further consideration: who will
>>> read this vector of pointclouds? What operations must be performed on
>>> it?
>>>
>>> Frequently, these problems are better solved by republishing some or
>>> all of the original points as a single PointCloud2 with a
>>> categorization field. The data can be sorted on that field for
>>> efficient access, sometimes that occurs naturally as a result of your
>>> algorithm, anyway.
>>> --
>>>  joq
>>> _______________________________________________
>>> ros-users mailing list
>>> ros-users@code.ros.org
>>> https://code.ros.org/mailman/listinfo/ros-users
>>>
>> _______________________________________________
>> ros-users mailing list
>> ros-users@code.ros.org
>> https://code.ros.org/mailman/listinfo/ros-users
>>
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>



-- 
MSc. Dejan Pangercic
PhD Student/Researcher
Intelligent Autonomous Systems Group
Technische Universität München
Telephone: +49 (89) 289-26908
E-Mail: dejan.pangercic@cs.tum.edu
WWW: http://ias.cs.tum.edu/people/pangercic